On Monday, November 16, the Commerce Department's Bureau of Industry and Security issued an interim final rule that requires licenses for advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil, and requires data centers to obtain a license before giving remote access to companies headquartered in China or other restricted destinations. It also formally rescinds the Biden administration's AI Diffusion Rule. Export control used to ask where a chip sits. It now asks who is logged in.
Two licenses and a threshold
The country piece is simple. On top of existing requirements, exporting, re-exporting or transferring advanced semiconductors to the five countries now needs a license. The rule cites a higher-than-normal risk of diversion there. Chips bound for companies headquartered in the US, or whose ultimate parent is, are exempt from this five-country license if they meet reporting requirements, which the rule sets monthly.
The remote-access piece is global. Per the rule's text, a license is needed to supply advanced computing items to any data center, anywhere, that the supplier knows will give companies headquartered in China, Macau or other Country Group D:5 destinations access to more than 12,000,000 aggregate TPP (total processing performance, BIS's compute yardstick), at one data center or across linked ones. Operators already serving such customers get 90 days from the effective date. So the 90 days is not a deadline for every operator, and its clock has not started: the rule takes effect on Federal Register publication, scheduled for Thursday, November 19, with 60 days of public comment after it. Commercial recommender systems, tools used only to rank, filter or suggest products, services or content, are exempt; generative AI is not.
Commerce said the rule ensures "existing restrictions on access to sensitive American technology cannot be circumvented", CBS News reported. The gap is familiar: some China-based firms barred from buying Nvidia's best chips have rented compute from Southeast Asian data centers instead.
A rescission, and an unusual byline
The AI Diffusion framework, issued by BIS on January 13, 2025, split the world into country allocations of TPP, with compliance not required until May 15, 2025. It was never enforced, and BIS said in May 2025 that it would rescind it. One idea survives the funeral: compute is still metered in aggregate TPP.
The byline is the odd part. BIS's release says the rule was drafted by, and at the direction of, the Treasury Department and the National Security Council. Chris McGuire of the Council on Foreign Relations said he had never seen a BIS rule that BIS did not write, per TNW.
The pressure was visible. On Friday, Treasury Secretary Scott Bessent faulted Commerce Secretary Howard Lutnick for not doing enough about chip purchases by Megaspeed, a Singapore cloud provider, and a day before the rule Sen. Maria Cantwell, D-Wash., wrote that "these companies are exploiting a loophole in U.S. export controls." Megaspeed, which US officials suspect of passing Nvidia chips to China, denies wrongdoing; TNW reports Jensen Huang quit as Nvidia's chief executive after revelations involving a former Megaspeed executive.
Who says what
McGuire called the rule "a major step forward", per CNBC, while noting loopholes remain, some inside the US. Joshua Kurlantzick, a CFR senior fellow, told Fortune the controls "could be disastrous" for Malaysia, and analysts said countries courting Chinese data center clients could be hurt.
Beijing's position predates the text. On November 13, asked about the reported measures, Foreign Ministry spokesperson Mao Ning said China "has consistently opposed the abuse of export control measures." Malaysia has since July 2025 required trade permits, with 30 days' notice, to export, transship or move in transit high-performance US-origin AI chips. We found no statement on the rule since Monday from the five governments, Nvidia, AMD, the large clouds or trade groups. Nvidia did announce today that its board has created a National Security and Export Control Oversight Committee, which does not mention the rule.
Nvidia fell as much as 2.9% on Monday and the Philadelphia Semiconductor Index as much as 5.5%, Bloomberg reported, per TNW. The Wall Street Journal reported, as relayed by Reuters, a plan to name more than two dozen AI "national champions", which Reuters could not immediately verify. President Trump meets Xi Jinping in Shenzhen on Thursday, and Nvidia reports results today, with its call at 2 p.m. Pacific.
What changes for a GPU renter
Johor's data center development pipeline reached 3,088 MW in H1 2026, against 1,110 MW operational; Bangkok's was 2,084 MW and Jakarta's 1,699 MW, per Cushman & Wakefield.

If your company is not headquartered in a restricted destination, the remote-access license mostly targets your provider's other tenants, but not only them: the rule's preamble also reaches a customer outside those destinations that, to the operator's knowledge, is training models for, or providing models to, a company headquartered in one. Our reading, not the rule's text: expect more ownership questions when buying GPU capacity in Singapore, Jakarta, Johor, Bangkok or São Paulo, since the operator now carries the risk of serving the wrong parent company. Chips bound for local operators in the five countries need a license, which could slow their new capacity. Teams in those regions for latency or data residency should keep a tested fallback region.
What we could not verify
- We read the rule's pre-publication text on the Federal Register's public-inspection list (document 2026-23529, 117 pages, RIN 0694-AK50), which can still change before its scheduled publication on November 19; BIS's own release could not be opened and is quoted here via Nextgov and CBS. The effective date is the publication date, and the 90-day and 60-day deadlines run from it, so none has a calendar date yet.
- Whether API access or managed inference counts is not settled by the text: it defines no "remote access" and never mentions APIs or inference, and turns instead on a supplier's "knowledge" of who the customer is and whom it works for. We have not seen that test applied.
- We did not map the rule's export classifications to specific Nvidia and AMD parts.
- Bessent's criticism and Cantwell's letter are Nextgov's alone; McGuire and Kurlantzick reach us via TNW. Errors would change who pressed and who warned, not the rule.
- Statements we missed from the five governments or industry would rebalance this. China's comment predates the rule; Malaysia's permits come from a July 2025 report.
- Market moves are Bloomberg intraday extremes via TNW, not closes. Reuters had not verified the WSJ champions report, context only.
- Cushman & Wakefield does not define "pipeline" or IT load, and reports metros: the chart shows scale, not national exposure.
- The Megaspeed allegations are unproven and denied; its link to Huang's resignation is TNW's alone. We found no reporting on European implications.
The 30,000-foot read
Borders used to be drawn at ports. This one is drawn at the login prompt, and the operator holds the pen, obliged to know whose parent company pays the invoice. Whether that is prudent security or overreach, Washington, Beijing and five other capitals will keep arguing; the Federal Register text will settle what it requires.
What an engineer controls is small and sufficient: where workloads can move, how fast, and whether your ownership chain is documented before anyone asks. Watch for publication, due Thursday, which starts both clocks; any guidance on what counts as remote access; any reply from the five governments; what Nvidia's management says tonight; and whether Thursday in Shenzhen changes the weather.
Sources: Federal Register public inspection, Addressing Remote Access to Advanced Semiconductors (interim final rule text, RIN 0694-AK50, scheduled for publication November 19: five countries, 12,000,000 aggregate TPP, Country Group D:5 and Macau, 90 days, 60-day comments, recommender-system exemption, US-headquarters exemption and its monthly reports, customers working for restricted companies; primary source); Nextgov/FCW, Commerce Department moves to close AI chip loophole (November 16 interim final rule, five countries, remote-access license for China-headquartered customers, 90 days from effective date, 60-day comments, rescission, Treasury and NSC authorship, Bessent criticism, Cantwell letter and quote, rental workaround); CBS News (export, re-export and transfer license for the five countries, Commerce rationale quote); The Next Web (12 million aggregate TPP threshold, D:5 scope, US-headquarters exemption with monthly reports, diversion-risk finding, effective on Federal Register publication, recommendation-system exemption, May 2025 rescission statement, Megaspeed denial and link to Huang's resignation, McGuire remarks partly via CNBC, Kurlantzick and analysts via Fortune, Bloomberg market moves, Trump-Xi meeting in Shenzhen); Federal Register, Framework for Artificial Intelligence Diffusion (BIS interim final rule of January 13, 2025: country allocations, compliance from May 15, 2025; primary source); Global Times (Mao Ning, November 13); Focus Malaysia (Malaysia's July 2025 permit regime and 30-day notice); Reuters via The Star (WSJ national champions report, unverified by Reuters); NVIDIA newsroom (November 17 call at 2 p.m. PT; primary source); NVIDIA newsroom, board oversight committee (November 17 release; does not mention the rule; primary source); Cushman & Wakefield, APAC Data Centre H1 2026 Update (pipeline and operational megawatts by metro). CAVEATS: rule details were checked against the pre-publication text of the interim final rule on the Federal Register's public-inspection list, which can still change before its scheduled publication on November 19; BIS's press release could not be opened and is quoted via Nextgov and CBS. McGuire's and Kurlantzick's remarks are relayed from CNBC and Fortune through TNW, and the market figures from Bloomberg through TNW. The Focus Malaysia article is dated July 15, 2025. China's statement predates the rule. Bloomberg's original report was not read. Scope note: the five-country license covers advanced chip exports to those countries generally, not only to data centers; the remote-access license applies to data centers anywhere; and the 90-day window applies to operators already serving restricted customers, counted from the effective date. Chart figures are Cushman & Wakefield's, by metro; Sydney and Mumbai are included for scale only.